Machine Learning-Based Nondestructive Determination of Apple Brix From Near-Infrared Spectra

Area of Science:

• Agricultural Science
• Spectroscopy
• Artificial Intelligence

Background:

• Apple quality assessment relies on Brix (sugar content) measurement.
• Nondestructive Brix determination methods often lack sufficient accuracy.

Purpose of the Study:

• To develop a highly accurate and efficient nondestructive method for determining apple Brix.
• To improve upon existing Brix prediction models using advanced algorithms.

Main Methods:

• A hybrid model combining Partial Least Squares (PLS) for dimensionality reduction and a Back-Propagation (BP) neural network for prediction was developed.
• An improved Genetic Algorithm (GA) was employed to optimize the initial weights and thresholds of the BP network.
• Spectral analysis was used for nondestructive data acquisition.

Main Results:

• The GA-optimized PLS-BP model achieved a high coefficient of determination (R²) of 0.9320.
• The model demonstrated a low root mean square error of prediction (RMSEP) of 0.2534.
• The model exhibited a fast inference time of only 1.2 ms.

Conclusions:

• The proposed GA-optimized PLS-BP model provides a fast, accurate, and reliable method for nondestructive apple Brix determination.
• This technique offers effective technical support for practical apple quality control.
• The study highlights the potential of integrating spectral analysis with optimized neural networks for agricultural product assessment.